Agents Lookup

Recipes / Operations

Supplier Purchase Order Follow-Up Agent

An agent scans your PO tracker for missing confirmations, ship dates, ASNs, and late orders, drafts supplier follow-ups for a buyer to approve, and proposes ETA updates only after a person checks the reply.

CategoryOperations
Steps10
Human sign-offs4

The problem

When a business buys from a dozen or more suppliers, purchase orders quietly go stale. A PO is sent, but no confirmation comes back. A supplier confirms but never provides a ship date. Goods ship without an advance ship notice, so receiving is caught off guard. Meanwhile the buyer is juggling sales, inventory, and cash, and the follow-up emails only get written when a stockout is already looming. The PO tracker, whether a Google Sheet, an Airtable base, or an ERP export, has the information needed to spot these gaps, but nobody has time to scan it every day. This recipe gives that scan to an agent, which drafts polite, specific follow-ups for the buyer to approve and send, then helps log supplier replies so ETA fields stay current, with a person confirming every change.

Who it's for

Not a fit when

Tools

PO tracker (Google Sheets, Airtable, or ERP CSV export)Source of open POs, expected dates, and status fields
Claude or another LLMDetects gaps against rules, drafts follow-ups, and extracts dates from supplier replies
Gmail or OutlookHolds drafted follow-ups for the buyer to review and send; source of supplier replies
Zapier or MakeRuns the daily scan and moves data between the tracker, agent, and email
SlackDaily summary of POs needing attention for the buyer
Google Sheets log tabRecords every follow-up sent and every proposed field change

Inputs you need first

Step-by-step build

  1. Standardize the PO tracker

    Make sure every open PO has one row with consistent columns: po_number, supplier_name, supplier_email, order_date, confirmation_date, promised_ship_date, asn_or_tracking, expected_arrival, status, last_follow_up_date, follow_up_count, notes. Use real date formats, not text. If you export from an ERP, set up a repeatable export to a fixed sheet tab so column positions never change. Messy dates are the most common reason this workflow produces wrong flags.

  2. Write the gap rules

    Define rules in a Rules tab so the buyer can adjust them without touching the prompt. Examples: no confirmation 3 business days after order_date; confirmed but no promised_ship_date after 5 business days; promised_ship_date passed with no ASN or tracking; expected_arrival within 7 days and status not shipped; last follow-up more than 4 business days ago with no reply. Compute these flags with sheet formulas so detection is deterministic and visible.

  3. Set contact limits

    Add columns for do_not_auto_draft and max_follow_ups. Mark strategic suppliers or those mid-negotiation as do_not_auto_draft so they only appear in the summary. Cap follow-ups at, say, three per issue before escalating to the buyer for a phone call. Record the supplier's preferred language and contact person so drafts are addressed correctly.

  4. Schedule the daily scan

    Use Zapier, Make, or Apps Script to run each weekday morning. Pull only flagged rows. For each, pass the agent the PO details, the specific flag, the follow-up count, and the last message in the thread if available. Post a Slack summary to the buyer listing flagged POs grouped by issue type, with a link to each row.

  5. Draft follow-up emails

    For each flagged PO not marked do_not_auto_draft, the agent writes a short, specific email: references the PO number and items, states exactly what is missing, asks one clear question, and proposes a reply-by date. The email is created as a draft in the buyer's mailbox, threaded on the original PO email when possible. Nothing is sent automatically. The agent never commits to new prices, quantities, or penalties.

  6. Buyer reviews and sends

    The buyer opens the drafts folder, edits as needed, and sends. A simple label or a column in the sheet marks each as sent, which updates last_follow_up_date and follow_up_count. Drafts not sent within two days are deleted from drafts and regenerated the next morning with fresh data so stale wording never goes out.

  7. Parse supplier replies

    When a supplier replies in a tracked thread, the agent reads the message and extracts proposed values: confirmation yes or no, promised ship date, tracking number, revised arrival date, and any problem mentioned such as partial shipment or price change. It writes these to a Proposed Updates tab with the source email link, never directly into the live tracker fields.

  8. Human check before updating ETA fields

    The buyer reviews the Proposed Updates tab, compares each extracted value against the email, and marks approve or reject. Approved rows are then copied into the live tracker by the automation. Anything involving price, quantity, or substitution changes is flagged for the buyer to handle personally, since those may need a revised PO.

  9. Test with past POs

    Before going live, run the scan on a copy of the tracker that includes several past POs you know were late or unconfirmed. Check that each expected flag appears, that no flag appears for healthy POs, and that drafts reference the correct PO numbers and dates. Paste in five real past supplier replies and confirm the extracted dates match what a person reads.

  10. Maintain weekly

    Review the log each week: suppliers with repeated follow-ups, POs that hit the follow-up cap, and extraction errors the buyer rejected. Adjust rules and lead times per supplier as patterns emerge. Refresh the do_not_auto_draft list when relationships change. Archive closed POs monthly so the scan stays fast and focused.

Example agent instructions

Adapt this to your business. Placeholders in braces are filled by your automation tool.

You assist the buyer at {business_name} with purchase order follow-ups. You draft emails for the buyer to review; you never send them. For the PO below, write a short, polite, specific email to {supplier_contact} at {supplier_name}. Reference PO {po_number} and the ordered items. State exactly what is missing based on the flag: {flag_description}. Ask one clear question and request a reply by {reply_by_date}. If this is follow-up number {follow_up_count} and it is above 1, acknowledge the earlier message without blame. Do not offer or accept changes to price, quantity, payment terms, or penalties. Do not invent dates, quantities, or tracking numbers. Write in {supplier_language}. Keep it under 120 words with a clear subject line that includes the PO number. Separately, when given a supplier reply, extract: confirmed (yes, no, unclear), promised_ship_date, tracking_number, revised_arrival_date, issues mentioned. Use 'unclear' or leave blank when the reply does not state a value explicitly. PO details: {po_row}. Previous thread: {thread_text}.

Human sign-off points

Guardrails

Expected time saved

Estimate: if a buyer manages around 40 open POs and spends about 2 to 3 hours a week scanning the tracker, writing follow-ups, and updating dates from replies, a daily flagged list with ready drafts might reduce that to roughly an hour of review and sending. Setup time depends on how clean your tracker is. Results depend on PO volume, supplier responsiveness, and data quality.

How to tell it's working

FAQ

Our ERP has its own PO module. Do we still need a spreadsheet?

Not necessarily. If your ERP can export open POs on a schedule or connect to Zapier or Make, the agent can work from that export. Many teams still use a sheet as a staging layer because it is easy to add rules, flags, and a Proposed Updates tab without changing ERP settings. Keep the ERP as the system of record and update it after human approval.

What if a supplier replies in a different language?

Language models can usually read and draft in many languages, but accuracy on dates and numbers matters most here. Keep the human check on every extracted value, record each supplier's preferred language in the tracker, and ask the buyer or a colleague who reads that language to review drafts for important suppliers until you trust the output.

Will automated follow-ups annoy our suppliers?

They should not, because every email is reviewed and sent by your buyer, is specific to one PO, and is capped in frequency. Suppliers generally respond better to a clear, single-question email referencing a PO number than to vague check-ins. Use the do-not-auto-draft list for relationships that need a more personal touch.

Want this set up for you?

Build it yourself with the steps above, or tell us what you need. We scope the work with you and send a written quote before anything starts.

Request a quote

Sources

Written by Agents Lookup. Official pages for the tools this recipe uses: